5250 Posted March 7 Posted March 7 (edited) While testing video joins problem, I discovered a workflow that makes the joins almost invisible. 35 seconds vides (1step is 5 seconds * 7 steps) Each videos SVIPro2Test.zip Edited March 8 by 5250 9
huehuehue9001 Posted March 7 Posted March 7 (edited) nice video. the transition is way better now. I found a new method. t2i illustrious > upscale illustrious > qwen > illustrious illustrious is still pretty rough, fingers, shoe lace, shoe, need fixing. qwen is actually pretty good at fine details and it fixed in one go. but qwen have this same face look. so put it back into illustrious to get the style. no need manual inpainting. lick it. when i slap it back in il, it ruined the hands. Spoiler >was browsing /sp/. >some model shot came up >thought it look like meg so here it is. added areola slip. the face is a bit off. didn't want to spend too much time on slop. Edited March 9 by huehuehue9001 8
yhb-q Posted March 10 Posted March 10 On 3/7/2026 at 9:40 PM, 5250 said: While testing video joins problem, I discovered a workflow that makes the joins almost invisible. 35 seconds vides (1step is 5 seconds * 7 steps) Each videos SVIPro2Test.zip 67.29 MB · 3 downloads 视频很漂亮，要是能达到1080规格就好了。
5250 Posted March 10 Posted March 10 (edited) 9 hours ago, 5250 said: The SVI Pro2 is amazing, so I experimented with many different workflows and combined them to create my ideal workflow. I stopped looping because I want to use a different LORA for each step. SVI has excellent consistency, but to ensure even greater consistency, I chose to specify an end image. I added that MMAudio option add to automatically generate the shutter sound when taking a photo. This is the movie made by this workflow. I mistake first step....(Neck choker) By use the various images that created in Forge into this workflow, I can create a realistic video too easy. AI is amazing. Nozomi.zip Edited March 10 by 5250 2
5250 Posted March 11 Posted March 11 (edited) 8 hours ago, huehuehue9001 said: I put it in qwen. The skin looks nicer. Hello. I tried to use Qwen. but I could not make good result... Maybe something wrong. So I checked T2V workflow to make beautiful pics. 4x Ultimate - Color Graded was the best. but it's need time to generate. Normal result (720 - 960) 4x-UltraSharp (2880 - 3840) 4x Ultimate - Color Graded (2880 - 3840) (* there is wrong point on foot) When compositing backgrounds with Forge, only one image could be generated at a time, so I created a workflow with ComfyUI to process multiple images simultaneously. DOAXVV_T2I_UltraUpscale.zip Edited March 11 by 5250
huehuehue9001 Posted March 11 Posted March 11 (edited) Since you like the realistic look, try to make qwen work. You can get nipple lora on civitai. After that adetailer the face. You can drag and drop my PNG into comfy and see. It's very similar workflow to sdxl base. You can't denoise too high, 0.5 max. It's the same as your nova asian. need to lower denoise, the skin is all wrinkly. lower until it's not blurry. and maybe lower cfg from 2 to 5. Lower cfg lower detail, but less error. cfg 1 is too blurry. Edited March 11 by huehuehue9001
5250 Posted March 11 Posted March 11 (edited) 9 hours ago, huehuehue9001 said: Since you like the realistic look, try to make qwen work. You can get nipple lora on civitai. After that adetailer the face. You can drag and drop my PNG into comfy and see. It's very similar workflow to sdxl base. You can't denoise too high, 0.5 max. It's the same as your nova asian. need to lower denoise, the skin is all wrinkly. lower until it's not blurry. and maybe lower cfg from 2 to 5. Lower cfg lower detail, but less error. cfg 1 is too blurry. Hello. Thank you for your advice. Could you compress(Zip file) the image and upload it? Because I can't get workflow from this thread Pics. I tried Qwen 2512 from template work flow. I think, Perhaps Lora for illustrious cannot be used in Qwen's environment. image_qwen_Image_2512.zip Edited March 11 by 5250
huehuehue9001 Posted March 11 Posted March 11 (edited) ComfyUI_00382_.zip You have to make a qwen lora. or t2i illustrious > qwen upscale > face detailer illustrious. Spoiler Edited March 12 by huehuehue9001 4
5250 Posted March 12 Posted March 12 (edited) 7 hours ago, huehuehue9001 said: ComfyUI_00382_.zip 5.38 MB · 1 download You have to make a qwen lora. or t2i illustrious > qwen upscale > face detailer illustrious. Reveal hidden contents Hello. Thanks for your reply. I see... So I can't use Loras for illustrious on Qwen. It's sad that I have a lot of Loras (about over 3000) for illustrious . Loras for Qwen is not yet enough, so I'll give up on switching to Qwen. Thank you. Edited March 12 by 5250
5250 Posted April 8 Posted April 8 (edited) Qwen3.5 TTS insufficient for creating new dialogue. Its support for Japanese, in particular, was poor, and it struggled with expressing emotions. However, the recently released "Irodoi TTS" is fantastic. It allows for very detailed specification of emotions. Now I can create any line of dialogue I want. However, MMAudio was still insufficient and could not reproduce the shutter sound cleanly. https://huggingface.co/Aratako/Irodori-TTS-500M-v2/blob/main/EMOJI_ANNOTATIONS.md 01.zip Edited April 21 by 5250 11
Recommended Posts
Create an account or sign in to comment
You need to be a member in order to leave a comment
Create an account
Sign up for a new account in our community. It's easy!
Register a new accountSign in
Already have an account? Sign in here.
Sign In Now